What gets automated is what people describe, not what they do
The written procedure and the real work rarely match. Automating processes on the first one breaks the second.
The MAPA method · AI process automationMap it first.
Then automate it.
Automate a process that already works, without breaking it.
The MAPA method is my approach to AI process automation: it starts with a good map. Before automating a process in your business, I map it as it works today: I interview your team with the help of an AI agent, find the exceptions nobody had written down and decide with you what to automate. Then I put it live in parallel with your team and keep watch once it runs.
You approve every phase. Data anonymised and kept in Europe.
One process at a timeParallel pilot before go-liveAnonymised data
Illustrative example · you approve every phase
Why automations fail
They don’t fail on the normal cases. They fail on the exceptions nobody mentioned.
Exceptions live in someone’s head
“Marta handles that customer”, “if they write in English, I pass it to Juan”. Nobody wrote it down, and that is exactly where automation breaks.
It goes live all at once
Without a parallel trial period, your customer is the one who finds the first error.
The MAPA method
Four phases. One rule in each.
The MAPA method is named after the initials of its four phases. It works as a cycle: if your process changes, it goes back to the first one.
- 01
Map
We understand and draw your process as it works today: interviews with management and the team, documents and, where possible, data from your systems.
You getThe process map, step by step, exceptions included
RuleNothing is automated without an approved map
- 02
Analyse
Step by step, we decide what to automate, what to support with AI and what to leave to a person, with the estimated savings and the risks.
You getDiagnosis and technical plan
RuleEvery decision comes with its reason and its data
- 03
Pilot
We build the automation with real cases from your business and run it in parallel with your team before switching it on in stages.
You getA tested automation and a trained team
RuleNothing goes live without a pilot that proves it
- 04
Audit
We watch results, new exceptions and changes to the process, with a monthly review. If the process changes, it goes back to M.
You getMonthly report
RuleNothing is left running unwatched
How it works inside
Fast AI detects. Conversational AI asks. The expert decides.
The agent combines two kinds of AI model, each used for what it does well, and reasoning that is never left to a model alone.
Fast intuition
A fast decision model
It doesn’t write text: it picks between closed options in milliseconds and says how sure it is of each answer.
What we use it for when mapping your process
- Checks every step of the map: is an owner, data or outcome missing?
- Spots contradictions between what two people say
- Proposes whether a step is automated, supported with AI or done by a person
When its confidence is low, it doesn’t guess: the question goes to a person.
Language
A conversational model
It understands and writes natural language, in the language of each person on your team.
What we use it for when mapping your process
- Interviews your team by chat, with clear questions
- Follows up when an answer doesn’t fit
- Pulls the steps out of emails, forms and documents
- Drafts the map, the diagnosis and the reports
It explains decisions; it doesn’t make them.
Judgement
The method and a human expert
Reasoning is not left to any model: it is written down, checked and approved by a person.
What we use it for when mapping your process
- Every fact on the map keeps its source and its confidence
- Fixed rules check that nothing is missing
- I review everything uncertain before it reaches you
Nothing moves to the next phase without your approval.
Privacy, anonymisation and security
Your processes say a lot about your business. We treat them that way.
Mapping a process means seeing how your team works and with what data. These are the baseline privacy and security safeguards on every project.
Anonymised before analysis
Names, emails, phone numbers and amounts are replaced with labels such as Customer A or Sales rep 2 before any model processes them. Understanding the process doesn’t require knowing who is who.
Only the data needed
We ask for the documents and access the process needs, read-only wherever possible. Nothing more.
In Europe, under GDPR
Data hosted in the European Union, a data processing agreement, and providers that don’t use your information to train their models.
Private area with controlled access
Your map and reports live in a private area, with your own login and two-step verification. No public links or loose files by email.
Everything is logged
Every fact on the map records who said it, when and from which source, and every change leaves a trail.
Least privilege
The automation can only do what was agreed: minimal permissions, keys kept in a secrets manager, and tests against manipulation attempts before go-live.
A person in charge
Decisions that affect customers or employees always go through a person, and we check what the EU AI Act requires in your case.
Deleted when we finish
We agree how long data is kept. When the project closes, it is deleted or returned to you.
The specific measures are set out in writing in the proposal, tailored to your process and your data. Guide: the EU AI Act and GDPR for SMEs
Which processes fit
Processes that repeat and depend on too many hands today.
- Handling enquiries and quotes
- Onboarding customers and suppliers
- Invoices, payments and reconciliation
- Orders, stock and purchasing
- Incidents and support
- Bookings and appointments
- Weekly recurring reports
- Documents and contracts
When it doesn’t fit
Processes that change every week, happen twice a year or depend entirely on one person’s judgement. In those cases I’ll suggest something else, or nothing.
If your process doesn’t need AI, I’ll tell you before we start.
After the map
Often, the result is a custom AI agent.
When a step needs to analyse data, prioritise or explain, the MAPA method ends in an internal custom AI agent designed on the map you already approved. And when a simple rule or integration is enough, I’ll tell you that too.
Guides
Further reading
Practical guides, with sources and examples, on the topics behind this page.
Frequently asked questions about the MAPA method
What is the MAPA method?
The MAPA method is how I automate business processes with AI without breaking what works. It has four phases: Map the process as it is today, Analyse what to automate and what not to, Pilot the automation in parallel with your team, and Audit it once it runs. An AI agent does the heavy lifting, and I review and approve every phase with you.
Why not just automate straight away?
Because the process described in a meeting is rarely the one people actually follow. Automations fail on the exceptions nobody mentioned; good process mapping costs little and avoids automating a mistake.
How much time does my team need to put in?
Not much, and no long meetings: management has one interview with me and the rest of the team answers by chat whenever they can. At the end we validate the map together in one session.
Does the AI make decisions on its own?
No. A fast decision model checks every step and, when it isn’t sure, asks instead of guessing; a conversational model interviews and explains. You and I make the decisions, and nothing moves on without your approval.
What happens to my customers’ and employees’ personal data?
It is anonymised before any model processes it: names, emails and phone numbers are replaced with labels. Data is hosted in the European Union, handled under GDPR and never used to train any AI.
What do I get at the end?
The map of your process as it is, a step-by-step diagnosis (automate, support with AI or leave to a person) with estimated savings, the technical plan, the automation tested in parallel and a monthly follow-up report.
What if the process changes later?
The audit phase catches it: when a new exception appears or a step changes, it goes back to the map, gets reviewed and the automation is adjusted.
How much does it cost?
It depends on the process and the tools. After our first conversation you get a fixed proposal by phase, and you can start with the map alone.